This is exactly the kind of story that sounds small until you sit with it for five minutes and realize it’s a preview of how information gets quietly poisoned.
OpenAI says it detected users in Iran generating around 100 fake articles with its chatbot, as part of an influence campaign that targeted at least 20 U.S. news outlets. That’s not a “wow, scary robots” headline to me. It’s a “we are about to drown in plausible junk” headline. And the most annoying part is how easy it is to shrug at the number. A hundred. So what. Twenty outlets. Big deal.
No. The point isn’t the count. The point is the workflow.
If you can produce something that looks like a normal article, in normal English, in the right format, and push it toward real media brands, you don’t need millions of pieces. You need the right few to land at the right time, in the right places. The goal isn’t to convince everyone. It’s to bend the atmosphere. To add fog. To make people argue, doubt, and split into camps. That’s what influence operations do when they work.
And let’s be honest: media systems are built for speed, not for careful proof. Even good outlets run hot. Editors juggle too much. Reporters are under pressure to publish fast. If you can feed that machine something that “sounds right,” you’re not trying to win a debate. You’re trying to slip something through a cracked door.
Imagine you’re a local editor and you get a clean, well-written piece tied to a breaking story. It cites “sources,” it has quotes that feel real, it uses the right tone. You don’t publish it as-is, but it shapes what you assign next. It nudges your framing. It eats time. Or imagine you’re not even a journalist. You’re a tired person scrolling, and a link is posted with a screenshot of a headline and a few paragraphs. You don’t click. You absorb. That’s enough.
This is why I don’t find it comforting when companies say, “We detected it.” Good. I’m glad they caught this batch. But detection is not the same as prevention, and it’s definitely not the same as resilience. It’s like celebrating that your smoke alarm works while your kitchen is still full of oily rags.
The uncomfortable truth is that generative AI doesn’t need to be perfect to be useful for manipulation. It only needs to be cheap, fast, and believable for long enough. A fake article doesn’t have to pass a courtroom test. It has to pass a vibe check in a group chat.
Here’s the part people will argue with me on: I think the bigger risk isn’t that Americans will be “fooled” into believing one specific lie. The bigger risk is that people will stop believing anything at all. When fake articles become normal, every real article becomes easier to dismiss. “That’s probably AI.” “That’s probably planted.” That’s not just confusion. That’s permission. Permission for bad actors at home and abroad to do whatever they want because the public can’t agree on what’s real.
And yes, state-linked operations are a real threat. But I don’t want us to pretend this is only a foreign problem. Once the technique is common, everyone uses it. Political groups. Scam networks. Random trolls with a grudge. The same tool that helps someone write a cover letter can help someone write a convincing lie at scale. That’s the trade. We get convenience, and we also get industrial-grade bullshit.
What should OpenAI do here? They’re in a tough spot. If they lock everything down, people complain that the tool is censored or useless. If they keep it wide open, they become a factory that others can rent for manipulation. And I don’t fully buy the idea that this can be solved with polite rules and “don’t do bad things” warnings. Influence campaigns don’t care about terms of service.
Still, I don’t think the answer is panic or banning AI. The answer is changing how we treat information. Newsrooms are going to have to assume they’re being targeted, even when the pitch looks clean. Readers are going to have to slow down, even when the post matches their politics. And platforms are going to have to stop acting like engagement is the same thing as truth.
The stakes aren’t abstract. If fake articles can reliably enter the bloodstream, then elections get noisier, public health gets harder, and international crises get more dangerous. Not because one fake story “wins,” but because the constant drip makes everyone more tribal and less willing to listen. The winners are the people who want chaos and cynicism. The losers are the rest of us trying to make decisions with half-broken signals.
What I don’t know is how many of these fake articles actually got published, or how close they came, or how many other campaigns weren’t caught. Public reporting rarely shows the full picture, and the people doing this aren’t going to post a scoreboard.
If we’re entering a world where producing believable text is basically free, what standard should we demand from news outlets and platforms before we trust what we’re reading?